A package for working with Azure Container Registry (ACR), Azure Kubernetes Service (AKS) and Azure Container Instances (ACI). Extends the Azure Resource Manager interface provided by the AzureRMR package.
AzureContainers lets you build and deploy containerised services in R, using Docker and Kubernetes. For full functionality, you should have Docker installed, as well as the kubectl and helm commandline tools. Otherwise it is relatively lightweight, requiring neither Powershell nor Python.
Note that AzureContainers can talk to any Docker registry that uses the V2 HTTP API, not just those created via ACR. Similarly, it can interface with Kubernetes clusters anywhere, not just those created via AKS.
The primary repo for this package is at https://github.com/Azure/AzureContainers; please submit issues and PRs there. It is also mirrored at the Cloudyr org at https://github.com/cloudyr/AzureContainers. You can install the development version of the package with devtools::install_github("Azure/AzureContainers").
Here is a sample R workflow to package up an R model as a container, deploy it to a Kubernetes cluster, and expose it as a service.
library(AzureContainers)
az<-AzureRMR::get_azure_login()
resgroup<-az$
get_subscription("<subscription_id>")$
create_resource_group("myresgroup", location="australiaeast")
# create container registryacr<-resgroup$create_acr("myacr", location="australiaeast")
# create Docker image from a predefined Dockerfile
call_docker("build -t newcontainer .")
# get registry endpoint, upload imagereg<-acr$get_docker_registry()
reg$push("newcontainer")
# create Kubernetes cluster with 2 nodesaks<-resgroup$create_aks("myakscluster",
location="australiaeast",
agent_pools=agent_pool("pool1", 2))
# give the cluster pull access to the registryacr$add_role_assignment(aks, "Acrpull")
# get cluster endpoint, deploy from ACR to AKS with predefined yaml definition fileclus<-aks$get_cluster()
clus$create("model1.yaml")
clus$get("service")